The Impact of HIV/AIDS on Current and Future Population Characteristics and Demographics in Botswana

Abt Associates Inc. The Forum 16 th Floor, 2 Maude Street P.O. Box 783400 Sandton, 2146 South Africa 27-11-883-7547 telephone 27-11-883-6790 facsimile 55 Wheeler Street Cambridge, Massachusetts USA 02138-1168 617 492-7100 telephone 617 492-5219 facsimile The Impact of HIV/AIDS on Current and Future Population Characteristics and Demographics in Botswana Study Two of the Investigation of the Socio-Economic Impact of HIV/AIDS In Botswana Hampden Square, Suite 600 4800 Montgomery Lane Bethesda, Maryland USA 20814-5341 FINAL REPORT May 2000 301 913-0500 telephone 301 652-3618 facsimile 640 North LaSalle Drive Suite 400 Chicago, Illinois USA 60610-3781 312 867-4000 telephone 312 867-4200 facsimile Prepared by: Abt Associates Inc. & Metropolitan Life Ltd. Commissioned by the Ministry of Finance and Development Planning with support from the United Nations Development Programme

TABLE OF CONTENTS EXECUTIVE SUMMARY 1 INTRODUCTION...7 2 CURRENT DATA ON HIV PREVALENCE, AIDS CASES AND MORTALITY...8 2.1 HIV seroprevalence and AIDS data...8 2.1.1 National HIV Sentinel Surveillance Survey... 8 2.1.2 Other HIV prevalence data... 11 2.1.3 AIDS data... 11 2.1.4 Other mortality data... 13 2.2 HIV/AIDS related behaviour and risk factors for the Botswana population... 14 2.2.1 Socio-economic and other factors affecting population risk of HIV infection 14 2.2.2 Information on risk behaviour and trends... 14 3 KEY DEMOGRAPHIC VARIABLES AND TRENDS... 16 3.1.1 Fertility... 16 3.1.2 Mortality... 16 3.1.3 Migration... 17 4 PREVIOUS HIV/AIDS IMPACT MODELING IN BOTSWANA... 18 4.1 Annual estimates of numbers of HIV infected people... 18 4.2 Projections using EpiModel... 18 4.3 Actuarial Society of Southern Africa Model (Version 500) projections... 19 4.4 Vensim... 19 5 REVIEW OF AVAILABLE PROJECTION MODELS... 20 6 METHODS AND ASSUMPTIONS IN DEMOGRAPHIC IMPACT PROJECTIONS... 23 6.1 Doyle Model Projections... 23 6.1.1 Scenarios... 23 6.1.2 Demographic input parameters... 24 6.1.3 Epidemiological parameters: macro-calibration... 26 6.1.4 Other epidemiological and clinical calibration parameters... 30 6.2 Spectrum projections... 31 6.2.1 Scenarios and HIV prevalence calibration... 31 6.2.2 Demographic parameters... 32 6.2.3 Epidemiological parameters... 34 7 RESULTS... 37 7.1 HIV/AIDS parameters.... 37 7.1.1 HIV prevalence... 37 7.1.2 AIDS cases... 37 7.1.3 AIDS deaths... 38 Abt Associates Inc. 2

7.1.4 Total population... 38 7.1.5 Population age profile... 39 7.1.6 Under-5 and infant mortality... 39 7.1.7 Life expectancy... 39 7.1.8 Births and birth rates... 40 7.1.9 AIDS orphans... 40 7.1.10 Dependency ratios... 40 7.1.11 Sex ratios... 40 7.1.12 Rates of new HIV infection... 40 8 INTERPRETATION OF PROJECTIONS... 41 8.1 General considerations... 41 8.2 Calibration of scenarios and sensitivities... 42 9 VALIDATION OF PROJECTIONS... 46 9.1 Empirical mortality data... 46 9.1.1 Family Health and Demographic and Health surveys... 46 9.1.2 Health statistics and Vital Registration pilot data... 47 9.1.3 Orphans... 48 9.1.4 Recent data from other countries... 48 10 CONCLUSIONS AND RECOMMENDATIONS... 50 10.1 Conclusions... 50 10.2 Recommendations... 50 10.2.1 HIV/AIDS data... 51 10.2.2 General mortality and fertility data... 51 10.2.3 Modeling and models... 52 10.2.4 Use of projections for planning and action... 52 11 NOTES... 53 GRAPHS OF PROJECTIONS ANNEXURE 1: MODELING DEMOGRAPHIC IMPACTS OF HIV/AIDS - MAIN MECHANISMS OF IMPACT ANNEXURE 2: AVAILABLE MODELS FOR PROJECTING HIV/AIDS IMPACTS ANNEXURE 3: SUMMARY TABLES OF DEMOGRAPHIC PROJECTIONS ANNEXURE 4: CALCULATION OF VARIOUS DEMOGRAPHIC AND EPIDEMIOLOGICAL PARAMETERS Abt Associates Inc. 3

EXECUTIVE SUMMARY Botswana has one of the most severe HIV/AIDS epidemics in sub-saharan Africa and the world. Women at 10 of the 14 Antenatal Survey sentinel sites had HIV infection levels of over 30% in the last two years, and all have levels over 20%. The epidemic has been hidden due to the long incubation period from HIV infection to onset of AIDS, and stigma that discourages disclosure of HIV/AIDS. However, it is clear that the epidemic is so severe that it will lead to dramatic changes in the population size and profile. Previous projections of demographic impacts in Botswana have used problematic assumptions around important micro-parameters and the shape of the epidemic curve for the country. Objectives The overall objective of this study is to provide Botswana with the best available projections of the demographic impact of HIV/AIDS on its population to 2010. Methodology A review of current HIV/AIDS data confirmed that antenatal survey provides the best available data to calibrate projections of the epidemic, and is likely to provide relatively good indications of levels and trends in infection rates despite a number of significant limitations. Other data of some usefulness includes reporting of AIDS cases to the AIDS/STD Unit, and data on behaviour and exposure to risk situations among Batswana. Demographic data in Botswana is relatively comprehensive and good quality. The 1991 Census is the main data source for making demographic projections. Family Health and Demographic and Health Surveys provide further estimates of fertility and mortality. Certain key parameters, particularly around adult mortality, are somewhat uncertain. Available models for projecting HIV/AIDS demographic impacts were reviewed. The study Technical Group requested that the consultants use the Spectrum Model to make final projections for Botswana. The Doyle Model was used to produce the HIV prevalence curve needed to calibrate Spectrum, and for other comparison projections. Scenario Assumptions Two main Spectrum HIV/AIDS scenarios (S1 and S2) were generated as a best estimate and best case scenario. A No-AIDS scenario (S0) provides a reference point. Key common assumptions and methods used in scenarios included the following. 1) A common antenatal seroprevalence epidemic curve for all Districts, with District epidemics differing only in their timing. 2) Calibration of projections against male and female age profiles of antenatal and reported AIDS case data. 3) Median adult survival times of 8.5 years from infection to AIDS, and one year after developing AIDS. Median child survival was assumed to be just under 2 years. 4) An average reduction of fertility of 34% among HIV-infected women. 5) Mother-to-child transmission rates of 30%. All scenarios add up projections for individual administrative districts to produce national figures. S2 assumes that rural populations have HIV infection levels consistently 20% below urban levels. Other scenarios (S3-10) show effects of interventions and sensitivity to changed assumptions. No projections incorporate migration trends. Findings In 1999 an estimated 15-17% of the total Botswana population were HIV-infected. Prevalence will potentially rise to 18-21.5% by 2005. Projected rates for adults aged 15-59 were much higher at 28-30% (240 000 adults), and could rise to 36% by 2005. Abt Associates Inc. 4

Prevention programmes, particularly among youth, will continue to be critically important in Botswana. Rates of new infection are projected to stay above 2% of the population per year until 2010 without effective prevention. Even fairly modest prevention targets could reduce adult prevalence by 5% or 40 000 people in 2010. The epidemic of AIDS cases and AIDS deaths is still at a relatively early stage. The 18 000-19 000 AIDS cases projected for 1999, will increase to 39 000-41 500 in 2005. An estimated 1.6-1.9% of adults in Botswana had AIDS in 1999. This will rise to 4.5-5% by 2010. Between 1.2 and 1.3% of adults were projected to die of AIDS in 1999, increasing to 3-3.5% of adults per year by 2010. A total of 327 000-350 000 Batswana are expected to have died of AIDS between 1991 and 2010, equivalent to 20-25% of the current population Preventing new adult HIV infections will have limited effect on AIDS cases and deaths before 2010, but longer-term benefits will be large. Programmes to reduce mother-to-child transmission could lower total AIDS deaths by about 6 % in 2010. Most deaths are expected to cluster in the 25-40 year age group. Men will be affected at slightly older ages than women. Botswana total population growth is declining rapidly due to HIV/AIDS. Only in the best case (S2) will growth remain positive (at 0.2%) by 2010. By 2010 the total population is projected to be between 1,53 million (S1) and 1.72 million (S2) HIV/AIDS will profoundly alter the population age profile. The greatest reduction in numbers compared to a no-aids scenario by 2010 will be among adults aged 35-45 (a 40%- 50% reduction) and children aged 0-9 years (32-40% lower). Under-5 and infant mortality will both increase. Under-5 mortality is expected to be 67-98/1000 higher than in a no-ads scenario by 2005. The IMR will be 20-25/1000 higher than in a no-aids scenario in 2000, rising to a 24-33/1000 increase by 2005.. Life expectancy is falling dramatically. It is projected to be only around half (46-52%) of no-aids scenario levels in 2010. Numbers of AIDS orphans will rise rapidly from an estimated 36 000-57 000 in 2000, 159 000-214 000 in 2010. Dependency ratios are expected to decrease slightly as numbers of new births will be impacted more than the size of the population aged 15-64. Interpretation of projections The projections given in this report are likely to give the best available indication of the scale and types of HIV/AIDS demographic impacts. However, no model can represent or predict reality perfectly. Projections are also very dependent on the quality of demographic and epidemiological data used for inputs and calibration. In using the projections the following general issues should be considered. All assumptions and methods used should be noted and critically reviewed regularly. Projections do not accommodate all potential non-aids factors, such as migration, that could change the demography of Botswana or districts. Aggregated, national projections are probably more reliable than district projections. Projections further into the future have more potential for error, but AIDS cases and deaths to 2010 are determined mainly by existing infections and will be fairly reliable. No scenarios except D3, S3 and D4 assume effective interventions. A vaccine or very effective new HIV/AIDS treatments are unlikely to be widely available before 2010. Many local communities will be hit much worse or less than indicated by projections. Projections indicate susceptibility to impacts and may not adequately reflect vulnerability of many districts, communities or groups to effects of death and illness. Abt Associates Inc. 5

Several specific issues should also be considered in interpreting the projections. General adult infection levels may be considered to be optimistic. Overall rates for people aged 15-49 average around 90% of antenatal estimates. Uncertainty exists about the appropriate values for several parameters in Botswana. These include: median survival times from HIV infection to death; fertility reductions among HIV infected women; rates of mother-to-child HIV transmission; realistic estimates of prevention programme effects; and the age distribution of infection. The S2 best case scenario may be too optimistic. Even if rural communities currently have lower levels of infection, they may reach high levels after a time lag. The Doyle model tends to project less severe impacts than Spectrum. Precise reasons for this are difficult to identify. In most areas however, the two models project very similar AIDS-related trends and magnitudes, and Doyle projections often reach similar values to Spectrum within 1-3 years. Many policy and planning decisions are likely to be fairly insensitive to which model is used. The 1996 FHS and 1998 DHS preliminary results do not show clear evidence of HIV/AIDS impacts on mortality, despite projected impacts that would be expected to be discernible. However, data from Health Statistics and Vital Registration pilots, while problematic, strongly suggest rising mortality. There is also widespread anecdotal evidence of rising death rates among young adult Batswana, and increasing evidence to support results of similar modeling in other countries. Sensitivities of projections to changes in several parameters are illustrated and result in relatively limited changes in magnitudes of impacts, which are unlikely to fully explain differences in survey and projection results. Declining underlying trends in mortality or inaccurate timing of epidemic curves in projections may be part of the explanation. Further scrutiny of DHS data and methodology may also reveal reasons for the apparent discrepancies. Conclusions Botswana antenatal survey data clearly indicates that the country faces HIV/AIDS demographic impacts of massive proportions. Projections are subject to many limitations but are likely to give a good indication of the size of various HIV/AIDS impacts. There should be no complacency about absence of clear HIV/AIDS impacts in the preliminary 1998 DHS results and 1996 BFHS. In so far as experience truly differs from projections so far, it is safest to assume that projections have inaccurately predicted the timing rather than the scale of impacts. Adequate information on the HIV/AIDS epidemic and its impacts must be ensured to inform planning. This will require devotion of greater resources and effort to both demographic and HIV/AIDS data collection. Several specific recommendations to improve data collection and modeling are made. In the interim, planning and action must commence to respond effectively to new and changing needs. All government sectors are likely to face substantial changes in the scale and types of needs to be met. HIV/AIDS impacts on employees will also reduce their capacity to promote development and meet specific HIV/AIDS-related need. Planning and implementation of responses cannot wait for perfect data or empirical confirmation of projections. Planners should be aware of the assumptions used in projections that they use in planning. They should incorporate risk analyses in to planning to identify whether low or higher impact projections will provide the lowest risk basis for specific plans. Abt Associates Inc. 6

1 INTRODUCTION Botswana is experiencing one of the most severe HIV/AIDS epidemics in sub-saharan Africa, the region which has the world's worst HIV/AIDS epidemic. The epidemic began to spread in Botswana during the 1980s. In the 1990s, HIV prevalence has risen dramatically. Since the first survey of HIV prevalence among pregnant women attending antenatal clinics was conducted in 1992, prevalence has climbed from 23.7% to 42.9% in Francistown and 14.9% to 39.1% in Gaborone. Other sentinel sites have also shown dramatic increases. Ten of the 14 sentinel sites have had a prevalence of over 30% in the last two years' surveys, and all sites have a prevalence of over 20%. Due to the long incubation period between HIV infection and onset of AIDS, the impact of AIDS among Batswana has only become more visible since the mid-1990s, and is still obscured by stigma and barriers to disclosure of HIV/AIDS status. The full impact of AIDS deaths and other impacts will only be felt around or after the end of the coming decade. However, it is clear that an HIV epidemic as severe as experienced in Botswana will lead to dramatic changes in population size and profile. The overall objective of this study is to provide Botswana with the best available projections of the demographic impact of HIV/AIDS on its population. Specific objectives included the following: Review current data on HIV prevalence, AIDS cases and AIDS related mortality Review key demographic variables and trends Review current assumptions used for HIV/AIDS projections in Botswana Recommend the most appropriate projection models for use in Botswana Demonstrate the magnitude of impacts using various models Identify the best available projections of current and future demographic impacts Identify potential behavioural changes which may alter the course of the epidemic and demographic projections The project team would like to acknowledge the valuable contributions of Ms T Botana, Mr B Molomo and other CSO staff, Dr D Veskov and Dr W Jimbo of the AIDS/STD Unit, Mr D Schneider, Dr R Greener, Dr W Sanderson, Ms G Beleme, and members of the Vital Registration Pilot project. Abt Associates Inc. 7

2 CURRENT DATA ON HIV PREVALENCE, AIDS CASES AND MORTALITY 2.1 HIV seroprevalence and AIDS data The main objectives of the HIV surveillance systems are to look at geographic spread of HIV, monitor trends over time, provide some indication of levels of infection in communities, and mobilise political support and plan for HIV/AIDS impacts. Several types of HIV/AIDS surveillance have been practiced in Botswana to date. 2.1.1 National HIV Sentinel Surveillance Survey The National HIV Sentinel Surveillance Survey has been conducted annually since 1992 among pregnant women attending public sector antenatal clinic services. The survey provides the best available calibration data for projection modeling. The survey has in most respects been conducted in accordance with WHO guidelines. Sample sizes have generally been considered to be adequate, and laboratory HIV testing standards have been found to be good. Antenatal survey trends and key features Table 1: Antenatal Surveillance - seroprevalence by site 1992-8 Site 1992 1993 1994 1995 1996 1997 1998 Francistown 23.7 34.2 29.7 39.6 43.1 42.9 42.96 Gaborone 14.9 19.2 27.8 28.7 31.4 34 39.08 Serowe/Palapye 19.9 29.9 34.4 Maun 12.7 19.4 33.1 33.53 Selibe Phikwe 27 33.1 49.89 Chobe 18.3 37.9 38.8 Lobatse 17.8 37.9 33.7 Southern 16 21.8 24.67 Kweneng 13.7 18.9 37.2 Tutume 23.1 30 37.45 Ghanzi 9.5 18.9 22.3 Antenatal data suggests several important trends in the Botswana epidemic. Prevalence levels have been above 33% in 9 of 11 sites included in the 1997 and 1998 surveys (Table 1). The other two sites are over 22%. These prevalences are as high or worse than those found in most regions of the worst affected countries in the world such as Malawi, Zimbabwe and Zambia. They are substantially higher than many other countries with more advanced epidemics such as Uganda, Tanzania and Kenya, where prevalence reached a plateau at lower levels. Only in Francistown, the leading site, has the epidemic shown repeated evidence of reaching a plateau. There is some indication that prevalence rates may be levelling off among the youngest age group in Francistown. This is often the first sign of success in reducing risk behaviour. However, sample size for youth tends to be small, which limits ability to identify such trends. Abt Associates Inc. 8

The epidemic prevalence curves generated in each survey site to date have a very similar pattern once allowance is made for time lags between the start of the epidemic in different areas. This suggests that epidemics as severe as those in the leading sites should be anticipated in all areas unless behaviour changes substantially (see Section 6). However, several uncertainties about the information provided by the Survey should be noted. 1 1) The Survey may not be fully representative of all pregnant women in Botswana. Several possible sources of bias exist, although their significance is not known. Several clinics are included per site, rather than a single clinic, as conventionally used in the WHO methodology. This is necessitated by Botswana's small population. Biases may be introduced because the survey does not cover private sector antenatal service users (some, mainly urban, women) or other pregnant women who do not attend public sector antenatal clinics. Different data sources give differing estimates of levels of coverage. Certain sites are not covered every year. This makes it more difficult to identify trends and patterns in epidemics in different communities at sub-national level. Data to rigorously track possible differences in HIV infection levels and trends between pregnant women in urban and rural communities is not available. Botswana appears to have rural prevalence that is unusually high relative to urban areas, when compared to some other countries. 2 Good communications infrastructure and highly mobile populations with oscillatory migratory patterns may explain this. Nevertheless, some difference is apparent and rural prevalence in North, South and Central Botswana have been estimated at an average of 83% of the urban prevalence in those areas. 1 3 However, there is still lack of clarity on the issue: Lower rural rates could reflect lower risk behaviour in rural populations and thus lower level epidemics in the long term. However, they may also simply reflect delayed development of epidemics in rural areas, a hypothesis made more plausible by the similar overall shapes of curves in all sites (see Section 6.1.3), or a combination of these factors. Only 1998 Survey raw data has been made available for analysis, giving no indication of trends Definitions of urban and rural areas used in the Census and the Surveillance Survey differ. Neither consider proximity to transport routes, which may be a stronger determinant of risk than urban-rural classification in itself. 2) Women in the general community may not be represented by Antenatal attenders. There are multiple, conflicting influences on this relationship. Their relative significance is unclear. Overestimation of community rates could occur if some lower risk behaviours (e.g. increased condom use, fewer partners) a are associated with lower pregnancy rates. This may be the case among teenagers, where Antenatal data may capture those with higher general rates of risk behaviour. Lower fertility among women in age a Family Health Surveys indicate that contraceptive rates for all methods rose from 24% in 1984 to 41.7% in 1996. In 1996, 11.3% of women reported using condoms compared to 1.4% in 1988, and 1% in 1984. Abt Associates Inc. 9

groups with lower prevalence (e.g. older women) would also lead to underestimation of community rates unless data is age standardised. Underestimation of community rates would occur if HIV infected women have lowered fertility. This has been suggested by several studies. 4 Lower fertility may be due to HIV infection itself and its complications, high STD prevalence and associated infertility, or high use of contraceptives other than condoms, as occurs in Botswana. Most recent studies suggest that Antenatal surveys tend to underestimate community prevalence. 5 The net effect of the various factors in Botswana is not clear, but it seems likely that Antenatal data could, if anything, underestimate community female prevalence. Antenatal Survey data may also have limitations in tracking population trends if various biases change over time (e.g. the impact of HIV on fertility increases with increasing average duration of infection). For demographic projection and planning purposes specifically, the following extra limitations need to be noted: The sampling approach used in the Survey is designed mainly to track trends in HIV prevalence in populations served by the sentinel sites, rather than to provide accurate estimates of levels on HIV prevalence in the Botswana population overall. The survey sites do not correspond ideally with Districts used for planning purposes, and for which Census data and population projections are usually reported. This limits ability to project HIV/AIDS impacts at these sub-national levels, which are sensitive to the local HIV prevalence. HIV prevalence data is not linked to variables such as education and employment status. More detailed analysis of data including health facility data, age and marital status is not performed routinely, even where sample sizes may allow for this. This may limit ability to identify current or emerging trends in sub-groups which may influence projections of impacts. It may also limit ability to target preventive interventions more effectively, evaluate interventions and identify likely social and economic consequences of HIV/AIDS. Recommendations Strong consideration should be given to the following. Community based surveys to assess the relationship between Antenatal prevalence and community prevalence, and possible changes over time in this relationship. Surveys could also assess impacts on fertility in HIV infected women, which has a major influence on demographic projections. Having a sampling system (Probability Proportional to Size) to obtain improved estimates of trends and absolute levels of infection nationwide. More frequent surveys in all sites that are currently not surveyed annually, to provide clearer information on levels and trends in infection, including whether epidemic curves in different areas differ substantially. Strengthening capacity to analyse Antenatal and other HIV data to allow for more detailed analysis, including trends in, for example, infection levels in rural and urban populations, and other sub-groups. Including other select data in survey data collection e.g. level of education. Abt Associates Inc. 10

Consideration should also be given to strategies to validate apparent declines in HIV infection rates once these begin to occur. In mature epidemics, AIDS deaths can result in declining prevalence while incidence rates are persistently high, even in young adults. 6 Sequential cross-sectional surveys of STD prevalence to monitor effectiveness of STD and HIV prevention strategies, may also be somewhat easier than studies of HIV prevalence or incidence. 2.1.2 Other HIV prevalence data HIV prevalence among STD patients Male sexually transmitted disease (STD) patients have been a sentinel population since 1992 and have documented very high levels of HIV prevalence among this group. Data from this source is of very limited value to guide assessment of HIV/AIDS impacts in the broader community. STD patients are not clearly representative of particular communities. The profile of clinic attenders may also change over time, making trends difficult to interpret. TB patients Some surveys of HIV prevalence among male and female TB patients have been conducted in Botswana. These can provide useful information to tackle the TB epidemic and understand its relationship with HIV/AIDS. However these surveys give little information that can be extrapolated from the patients surveyed to more general populations, and thus are not likely to be helpful to calibrate demographic projections. Blood donation screening Blood donations in Botswana are screened for HIV to ensure safety of blood supplies. Data from this source is however of very limited use in surveillance. Blood donors are not clearly representative of the broader community or particular sub-groups, and levels of infection are not reliably similar to the general population. 7 Biases are introduced by factors that include efforts to reduce donations by people at high HIV risk, and selfselection among donors, including using donation as a means of testing, which has been reported anecdotally in several instances in Botswana and South Africa. These biases may also change over time, so that not only levels of infection, but also trends, are difficult to extrapolate to broader populations. 2.1.3 AIDS data Information on AIDS and AIDS deaths is less important than information on HIV infection rates to monitor the epidemic and impact of interventions. However, the data is potentially useful for advocacy purposes, and to validate various assumptions made in projecting impacts of the epidemic. Data on AIDS cases, HIV cases/carriers and AIDS deaths are potentially available from three main sources: Abt Associates Inc. 11

Data collected by the AIDS/STD Unit, linked to laboratory requests for testing for HIV. Reports from hospitals, primary care services and on deaths outside of institutions, which are reported in the Health Statistics documents. Vital Events data, produced annually by the Civil Registration system. All of these sources of data suffer from shortcomings. The most important is underreporting. A significant number of AIDS cases and deaths may not come to the attention of health facilities or be reported. In 1996 3028 deaths were reported by Civil Registration compared to 1480 in Health Statistics. Furthermore, only cases with confirmed laboratory test diagnoses are reported. ASU estimates that around half of people with clinical indications for HIV testing refuse to be tested, a proportion supported by responses of hospitals surveyed as part of the health sector impact study. Patients at facilities staffed by nurses only, who cannot draw blood, and at private facilities, are also under-represented. Figure 1: AIDS cases reported to the ASU b 8000 7000 6000 Cases 5000 4000 3000 2000 1000 0 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 New Cumulative In addition, the reported cases may not be a representative sample of people in the population with AIDS/HIV. This means that the data may not give an accurate reflection of e.g. age and gender profiles of AIDS cases. Specific biases may include: Double counting, recognised as a problem by the AIDS/STD Unit. Possible, but untested, differences in health care seeking and test acceptance behaviour between different groups (e.g. age, sex, occupation). Anecdotal reports by some health service staff suggested, for example, that women tend to present earlier in the course of HIV disease than men and often leave hospitals without HIV testing, with a greater tendency to remain at home in the terminal stages of their illness. Many men on the other hand, were reported to often have very limited family ties and support systems. This meant that they tended to stay longer in hospital with late stage illness, at which time they were more likely to agree to HIV testing. b Source: AIDS update April 1998, HIV/STD Unit, Ministry of Health Abt Associates Inc. 12

Studies and anecdotal evidence from elsewhere indicates that people with late stage disease may increasingly not present at health facilities. This may be because of factors such as increased crowding in facilities, emerging alternative care strategies or lower optimism about benefits of medical care among people with AIDS. 8 2.1.4 Other mortality data Other sources of mortality data, without a diagnostic link, include the Census, and the Family Health Surveys (conducted 8-10 yearly), and the Demographic and Health Surveys (10 yearly) which are conducted between Census'. These may be able to pick up trends in mortality that can be ascribed to AIDS in inter-censal periods. A recent review of mortality trends in sub-saharan African countries' household surveys and census' has found that these can be used effectively to monitor HIV mortality impacts. 9 However, difficulties in interpreting trends, or lack of expected trends, in infant and childhood mortality have been noted in other countries even with very advanced and severe epidemics, where other trends appear to mask the influence of HIV. c 10 The 1991 Census was too early in the course of the epidemic to be expected to identify any HIV related trends. In addition, the Census questions in relation to adult mortality were somewhat ambiguous in definition of the household, leading to difficulty in interpreting adult mortality data. The 1996 FHS preliminary analyses do not show an obvious impact on infant or child mortality: infant mortality declined from 39.5/1000 (1978-1988) to 39/1000 (1986-1996). Childhood mortality rates declined from 17/1000 (1978-1988) to 10/1000 (1986-1996). Analyses of the 1998 DHS is not yet complete. Preliminary analyses report no clear increase in adult, infant and child mortality. Use of existing Civil Registration data is limited by incomplete coverage of the country (in certain areas registration has not been compulsory), under-reporting at the time that events occur, and delays in submission of data in certain areas. Significant biases in the Registration data of relevance to HIV/AIDS projections cannot be excluded. Four districts have participated in a pilot project to increase coverage and reliability of vital statistics registration. These show strong evidence of rising mortality in the two larger pilots but it is not completely clear what biases may influence this data. The forthcoming Census should provide key data sufficiently far into the AIDS epidemic to provide good indications of mortality levels and trends. Recommendations Mortality data that is not linked to HIV/AIDS specifically will be a cornerstone of monitoring HIV/AIDS impacts and validating projections as the epidemic progresses. Fertility trends, which are measured in many of the same surveys or mechanisms, will also be important to monitor as many determinants of fertility cannot be predicted easily in a population heavily affected by HIV/AIDS. Adequately frequent, timeous and reliable mortality data for adults and children must be available. Capacity should be assured so that surveys can be performed rigorously and analysed quickly. c See Section 9 for further discussion. Abt Associates Inc. 13

Vital Registration pilots can provide useful information. Potential biases of data from the pilots, in the absence of a more general, improved system of reporting which helps to clarify denominator populations in particular, may however make it difficult to interpret their data. Given the importance of the issue for planning, there may be justification for increasing the frequency of large surveys such as the FHS or DHS. Alternatively, more specific surveys or sentinel sites may be required to provide more detailed and frequent information in interim years to identify and track trends. This is particularly important to consider as the AIDS epidemic is now in a rapid growth phase when AIDS deaths will become much more discernable in standard surveys and surveillance systems. All methods should be scrutinised to avoid possible biases introduced by the nature and scale of the HIV/AIDS epidemic. For example, some methods may not pick up data relating to households that dissolve due to death of adults. This factor is likely to become increasingly prominent as adult deaths will occur on a much larger scale than before, and AIDS deaths will tend to cluster among adults in the same household. 2.2 HIV/AIDS related behaviour and risk factors for the Botswana population For calibration of certain HIV/AIDS demographic models, and to inform estimates of future progress of the epidemic, two broad sets of information are relevant. 2.2.1 Socio-economic and other factors affecting population risk of HIV infection The Botswana population has several underlying features that suggest particularly high risk of a severe epidemic: High levels of mobility and good infrastructure. A tradition of high levels of movement of the population, particularly men, between their different homes in urban areas, villages, lands where crops are planted and cattle posts. High levels of STDs other than HIV 11. High levels of unemployment, particularly among youth 12. Poverty and inequality 13. Low status and poverty of women, which makes it more difficult for them to protect themselves from infection. 14 Low rates of marriage, which may predispose to less stable sexual partnerships. 15 Reportedly high rates of alcohol abuse, which can make people less careful or able to avoid high risk sexual behaviour, and may occur in social environments with more access to high risk sex. These factors are likely to predispose Batswana to the high rates of casual sex and partner change reported among them (see below). 2.2.2 Information on risk behaviour and trends Several Knowledge, Attitude and Practice (KAP) studies have examined risk behaviour among young Batswana. 16 Some changes in risk related knowledge, attitudes and behaviour/practice among young Batswana do seem to be occurring. Abt Associates Inc. 14

The degree to which reported attitudes and practices reflect consistent behaviour or behaviour change is unclear. Sampling differences and other methodological issues limit ability to compare studies conducted in different years. However, there is evidence of improved knowledge, attitudes and practice which may lead to reduced infection rates and affect the course of the epidemic. In 1996, around 80% of a sample of rural and urban youth knew of at least two valid ways to protect themselves from HIV infection. This level of information is higher than found in women and men in the 15-59 and 30-34 year age groups in Malawi in 1996, when there was little evidence of decline in youth HIV incidence. 17 The percentage of sampled sexually active people who had casual partners was 75% in the 1994 samples and 50% in 1996. Condoms have become more acceptable as a form of protection against HIV/AIDS. Condom use in casual relationships during the last sexual encounter reportedly was around thirty percent in 1992 and 85% in 1996. Condom use in regular relationships was over 50% in 1996. However, condom use in regular and casual relationships still seems to be inconsistent. The 1996 FHS also suggests significant increases in condom use, and that it accounts for over 50% of reported contraceptive use in 15-19 year olds. Abt Associates Inc. 15

3 KEY DEMOGRAPHIC VARIABLES AND TRENDS Botswana has shown marked changes in demographic variables and trends in previous decades. For population projections, the 1991 Census will remain the main data source for the foreseeable future. Botswana Census data is generally considered to be reliable, and the "gold standard" for demographic use. Census data provides information broken down for populations of districts, the major urban areas, urban villages and rural communities. No substantial deviations from previous fertility and mortality trends have been reported by CSO staff in preparation of BFHS III preliminary results. As a sampling approach is used in the BFHS and the focus of the survey is somewhat different, there are inherent limitations on the ability of such surveys to provide a reliable substitute for Census information on all key demographic parameters, although the data derived from them is still informative. Several key, recent demographic features of Botswana are critical to projecting HIV/AIDS impacts. (Further details of demographic parameters are provided in sections on methodologies used in projecting HIV/AIDS impacts below). 3.1.1 Fertility Fertility in Botswana has shown marked declines over recent decades. Total fertility rates in both Census and each BFHS have shown persistent reductions since the early 1980s, from over 7 in 1980 to an estimated 4.2 in preliminary 1996 BFHS results (Table 2). Fertility has persistently been highest in women aged 20-24 and 25-29 years. Table 2: Total fertility rates among women a) Census 1981 1991 TFR 7.1 4.23 b) Botswana Family Health Surveys 1984 1988 1996 TFR 6.5 5 4.2 c) Adegboyega 1998 1981 1991 TFR 7.07 4.73 3.1.2 Mortality Botswana has experienced a substantial decline in childhood mortality rates in recent decades, although trends over the 1990s await further information from final BFHS and DHS results. Data on adult mortality is considered to be less reliable than for children. The relevant questions in the 1991 Census are thought to have led to double counting of adult deaths. Current best estimates of mortality profiles are thought to be those Abt Associates Inc. 16

produced by E. Udjo in Population and Housing Census Analytic Report CSO 1998, but require confirmation in further surveys. Various published estimates of key mortality indicators are given in Table 3. Table 3: Published mortality estimates for Botswana 1981 * 1988** 1991*** 1991 * 1996** 1996* IMR 85.9m; 39.5 38 48 39 42 65.1f (51m; 27f) CMR 17 53 m; 27f 10 Under-5 MR 52.7 49 Life expectancy 57 m; 65 f 63m; 67f 69.9 64.8m; 68.4f CDR 11.5 (13.4m; 9.5f) 7.7 6.6 Sources: * Population Projections (1997); ** Family Health Surveys (1988; 1996 provis); ***Udjo (1998) 3.1.3 Migration International migration is not expected to be a major influence on demographic profile or the effects of the HIV/AIDS epidemic in Botswana going forward. Only 2.2% of the population were non-citizens in 1991. This proportion is expected to decline under current projections. Declining migration of Batswana to South African mines is expected to be offset by increased migration of citizens due to socio-political changes in the region. Urban population growth is not expected to be as fast as in the 1980s, when it averaged 13% per annum, largely due to reclassification of many large villages as urban areas. 1991 Census data suggests that urban-urban migration has tended to be the main type of migration (34.4%), followed by rural-urban (25.5%) and urban rural (20.9%). Abt Associates Inc. 17

4 PREVIOUS HIV/AIDS IMPACT MODELING IN BOTSWANA 4.1 Annual estimates of numbers of HIV infected people Previous Sentinel Survey reports have estimated the number of infected adults and children in Botswana in the given year, based on estimates for Gaborone, Francistown, other urban and rural areas. Several important assumptions are used in these estimates. 1. Antenatal prevalence equals prevalence in the adult female population aged 15-49. This assumption is defensible in the absence of population survey data, although age standardisation may have improved estimates, all else being equal. 2. Male:female prevalence ratio. This is assumed to be 0.83:1, which seems reasonable in the absence of a population based survey to validate the figure. Population-based studies provide limited and divergent information on expected male:female ratios, for example, 0.97:1 (95%C.I. 0.70:1; 1.35:1) in Ethiopia, and 0.68:1 (95% C.I. 0.59:1; 0.73:1) for urban and 0.89:1 (0.87:1; 0.90:1) for rural areas in Zambia. 18 A recent UNAIDS review of 15 community seroprevalence studies in Africa suggests that on average the ratio of infected men to women in Africa is in the region of between 0.77:1 and 0.83:1. 19 The ratio differs with age, and age standardisation is required for rigorous comparisons. In addition, the ratio tends to change according to the stage of the epidemic. Typically differences are greater between men and women in younger age groups, and prevalence in older age groups (>30 years) becomes more similar for men and women, or may become higher in men. 3. Rural prevalence. This has generally been estimated at around 50% of urban areas, an assumption which appears likely to significantly underestimate rural prevalence (see above). 4. Assumptions of prevalence over the age of 50. This is assumed to be 1%, a figure with no substantial justification and possibly an underestimate (projections suggest 1.8% in 1999). However, this is unlikely to have biased estimates to a large extent, in view of the low prevalence and relatively small proportion of the adult population aged over 50. 5. Infections in children aged 0-4. These are reportedly estimated as 15% of adult infections, based on the assumption that for every six infected adults (3 partnerships) there would be one infected child. This figure is clearly problematic and the estimated 37000 infected children in 1998, is high relative to other estimates based on current knowledge of mother-to-child transmission and mortality rates. Three models have previously been used for HIV/AIDS projections in Botswana. The major features and limitations of these models are reported in Annexure 1. 4.2 Projections using EpiModel EpiModel has been used for the population projections made for Botswana in 1997, and by the AIDS/STD Unit to produce projections included in various reports on the HIV/AIDS epidemic in Botswana. The model has been used extensively by UNAIDS in projections for Botswana and other countries. Previous population projections in 1997 incorporated several important assumptions. The epidemic is assumed to have begun in 1983. Abt Associates Inc. 18

The estimate of median survival time to AIDS is 10 years, with 100% mortality 2 years after development of AIDS. The peak incidence of HIV is assumed to occur one year from the reference year. The vertical transmission rate is 33% Antenatal prevalence is representative of the general population in sexually active age groups Rural HIV prevalence is 50-100% of that in urban areas for high and low estimates. The ratio of infected women to men is 1:0.83 Three HIV/AIDS epidemic scenarios were projected in the 1997 Population Projections: The Low Variant assumed a gradual decline in prevalence of 2% per annum from an estimated adult prevalence in 1991 of 10.49%. d A Medium Variant assuming a constant 10.49% adult prevalence for the duration A High Variant assuming a 5.2% annual increase in prevalence until a plateau at prevalence of 50% in the 15-49 year age group. In general, previous projections using Epimodel have been made with assumptions which HSU and CSO staff involved in modeling feel are too arbitrary, particularly in relation to the position of any calibration point on the epidemic curve and the shape of the curve. The Low and Medium variant prevalence assumptions are, with hind-site, too low. In addition, the model can only be used for short-term projections, and provides no substantial differentiation of impacts in terms of age group or sex. 4.3 Actuarial Society of Southern Africa Model (Version 500) projections This has been used by the Botswana Insurance company. Various aspects of calibration have been changed to reflect the experience of Botswana, and the model has reportedly performed adequately in relation to claims experience of the insurer. Details of the projections and claims experience are not in the public domain and overall demographic projections for Botswana have not been the focus of the modeling by the company. However, the company has indicated that when compared to calibrations of the model based mainly on South African data, their Botswana calibrations suggest: Earlier commencement and more sustained levels of sexual activity Overall higher risk behaviour (numbers of partners and levels of STDs) Potentially higher levels of infectivity due to cofactors such as other STDs. 20 4.4 Vensim Vensim has been undergoing final adaptation and calibration by Dr Warren Sanderson, as part of a broader study of environmental and demographic factors in development in Southern Africa. The robustness of projections and technical aspects of the model and its calibration are not available for close scrutiny. Certain initial projections (specifically HIV prevalence trends) seem out of line with experience in other epidemics and the reasons for this are not clear. d This appears to have been a mainly arbitrary adjustment to an estimated overall Antenatal prevalence of 14% in 1991. Abt Associates Inc. 19

5 REVIEW OF AVAILABLE PROJECTION MODELS Seven widely recognised models were reviewed for potential use by Botswana for HIV/AIDS projections, and comments on several others were made (see Annexure 1 for further details). A summary of the features of these models is provided in Table 4. While the review was not exhaustive, it provided a sufficient range of approaches to modeling the Botswana epidemic to allow for informed choices of model(s) for further work. Several overall issues in choosing the most appropriate model for use in Botswana were identified. As the real world is complex, no model will be able to simulate it perfectly. All models involve many assumptions. It is possible that two different models can be equally successful in describing reality. The superiority of any particular model is difficult or impossible to establish conclusively. Some good models of reality, based on solid theory, may perform less well if supporting data and tools are inferior. The outputs of any model and tool are highly dependent on available input data, and can impose greater limitations on projections than the logic of models themselves. Whatever model is chosen, it is likely to be desirable to establish the sensitivity of projections to key assumptions. Major criteria in making final choices between models included: The needs of the user. In general, models are designed to meet a specific need. A model should be chosen which is able to produce the relevant, reliable projections for planning and advocacy purposes in Botswana. Availability of calibration data and ability to use all relevant data as fully as possible. In Botswana, where antenatal data is probably the best available input, a macro model has strong advantages. However it is also important to have sufficient micro parameters to reflect epidemic characteristics which may be unique to Botswana, allow study of potential interventions and provide a stronger basis for longer-term projections. Sufficiently detailed outputs that can be tested against available data are also desirable to validate projections. Ongoing maintenance and technical support. Current data on the Botswana and other sub-saharan epidemics is limited. Projections and calibrations will probably need to be revised as new data becomes available. Unless relatively simple models are chosen, there might be constraints on availability of staff to maintain and update the model within the Botswana government. After reviewing these and other considerations (see e.g. Table 4) the study Technical Group opted to rely on Spectrum as the main model for projections in the immediate future. The Group suggested that the Doyle model be used to generate plausible epidemic curves for calibration of Spectrum projections and comparison projections. Abt Associates Inc. 20

Table 4: Summary of performance of models in terms of set criteria *** Strongly compliant, * Weak compliance,? Some uncertainty, No entry in cell = non-compliance with criterion Criterion 1. Input/calibration data Epimodel Spectrum IwgAIDS SimulAIDS Vensim ASSA Doyle Population profile (by age, gender) *** *** ** *** *** *** Fertility rates (by age, change over time) *** *** * *** *** *** Impact of HIV on fertility *? * *** ** Non-AIDS mortality rates (age, gender, over time) *** *** * ** *** ** Antenatal HIV prevalence data * *? **? *** *** AIDS cases AIDS deaths STD patient HIV prevalence * ** ** ** **? * *** * *** ** **? *** *** ** * *** *** Consideration + pooling of sub-group/ region data *** Medical and behavioural interventions * * *** ** * * ** Availability of reasonable input data for Botswana *** *** ** ** ** Ease of calibration * * *? ** ** 2. Outputs HIV prevalence (by gender, age) * ** *** *** **? *** *** HIV incidence (by gender, age) * * *** * *? ** *** AIDS cases (by gender, age) * ** *** * ***? *** AIDS deaths (by gender, age) * ** *** ***? *** *** Outputs by sub-group (e.g. district) * *** Infant, child, young adult mortality ** *** **? *** *** Abt Associates Inc. 21